topic RE: Round-Robin Trunking in ExtremeSwitching (EXOS/Switch Engine)

Round-Robin Trunking

Mrxlazuardin — Thu, 16 Apr 2015 17:07:00 GMT

Hi,

Is there anyway to configure round-robin algorithm on trunking (ports grouping)? I can do that easily on Linux, but I don't have idea on EXOS. With that algorithm, the load of each member of ports group will be balanced, no matter of the source and destination.

Basically it will works if the each member doing Tx in round-robin manner and all members doing Rx.

Best regards,

RE: Round-Robin Trunking

jackmikel — Thu, 16 Apr 2015 17:19:00 GMT

Maybe lacp can take care of it:
e.g.
ena sharing 1 grouping 1-2 lacp

RE: Round-Robin Trunking

Mrxlazuardin — Thu, 16 Apr 2015 17:19:00 GMT

Hi Jack,

Based on my experience, LACP is not so "balanced" on port grouping of connection to server or any end node. It is because of the destination is same.

Best regards,

RE: Round-Robin Trunking

Paul_Russo — Thu, 16 Apr 2015 18:32:00 GMT

Hello Mrxlazuardin

We do not support round robin on LAG ports as it can lead to packet out of sequence errors.

Hope that helps

P

RE: Round-Robin Trunking

Mrxlazuardin — Thu, 16 Apr 2015 18:32:00 GMT

Hi Paul,

Don't you think that unsequenced packet is common in TCP/IP networking and TCP/IP has capability to deal with that?

Best regards,

RE: Round-Robin Trunking

dflouret — Thu, 16 Apr 2015 18:32:00 GMT

Mrxlazuardin,

TCP does indeed have the capability to deal with out of order packets. But to reorder them requires storing misordered packets in buffers which requires memory and increases latency.

In a slow link, a few misordered packets don't hurt. In a LAG of high speed links the number can be high and so will be the amount of buffer memory and the processing time needed to reorder them.

RE: Round-Robin Trunking

dflouret — Thu, 16 Apr 2015 21:07:00 GMT

No. EXOS supports only port-based or address-based algorithms:

RE: Round-Robin Trunking

Paul_Russo — Thu, 16 Apr 2015 23:34:00 GMT

Mrxlazuardin I would hope that your network doesn't have a lot of out of sequence packets. Yes they do happen but it is better to send a whole flow over the same link to min any disruption. In a situation where you have one hop maybe there is less concern but across a network that is fro Edge to aggregation to core having a lot of that traffic could be a big issue.

Because our products can go from Enterprise to SP, from k-12 to high end DC or internet exchanges we need to have confidence that the network will work in all environments.

Thanks
P